Local interaction evolution strategies for design optimization

This paper studies and describes two classes of local interaction evolution strategies (LES) where the choice of parents for mating as well as the decision for survival and replacement are performed locally only. The entire population is now considered to consist of spatially distributed individuals each belonging to its own overlapping neighborhood and selection operates on a set of local pools. The specific real-world application considered is the optimization of a 2-lens-system being part of a heterodyne receiver, a micro-optical communication module. The goal is to optimize the system to be as insensitive to fabrication tolerances as possible while still maintaining optimal properties. This task is highly multi-modal and multi-objective too. The evaluation of each design variant is based on a simulation model and thus is a time consuming task.